Original article was published on Deep Learning on Medium
Hype that surrounds machine learning
Machine learning is an coalescence of different fields, which includes computational statistics, artificial intelligence, pattern recognition and mathematical optimization, among others. There is a misconception that only a polyhistor could become a Machine Learning expert, but still everyone who we know or meet is a machine learner.How is that even possible?
Surprisingly, Machine learning is not a new topic, its root can be traced back to the 17th century, with introduction of Baye’s Theorem by Thomas Baye’s. while the first neural network was developed in 1951, RNNs in 1982 and reinforcement learning in 1989. So i believe its necessary to know how to navigate through the hype.
Firstly for any new technology its necessary to know its strengths and weaknesses. for this Research papers might be helpful, but it will only provide results for the best cases, then draw the conclusion for a more generic setting. for example deep-learning methods have shown improvement in the speech and image processing, a similar inference can’t be drawn for text-mining.As a fact, associating semantic meaning to text and predicting moving target like flu still remains challenging.
secondly, finding resources that can possibly act as a bottleneck is very important.Many methods and algorithms were developed during 1960’s ,but due to computational limitations they were not applied in the real applications.
Finally, we should focus on understanding the underlying technical implementation rather than just continuing to use new technologies . For example, complex multi-layer deep learning models can be built in Keras with just under seven lines of code, which significantly reduces the barriers to entry. However, it is important to know how to solve problems and improve the performance of the model.
In conclusion, gaining all this knowledge will naturally help us to cut hype.mining reality out of hype is a complicated process with no simple solution. When people give in to hype and draw faulty inferences, it is alarming. However, the limitations that are omitted during these applications are the inherent bias in machine learning models that could come from a bias training set.